Abstract

Retirement age is often used as a proxy for retirement leisure, but if retirement is correlated with mortality this may be misleading. Using data from the Health and Retirement Study and an ordered SUR Tobit model, we analyze the determinants of retirement and death age to see who consumes retirement leisure. We find that men, Hispanics, white collar workers, people in good health, and workers with defined contribution pensions or high defined benefit accruals consume less retirement leisure. We also find a variety of factors that significantly influence retirement independently but do not affect retirement leisure, resulting in misleading predictions.

Recent discussions concerning public and private pension plans and their rules have turned the institution of retirement into a highly contentious political and economic issue. One of the primary controversies turns on the perceived lack of sustainability of traditional defined benefit (DB) pension plans. An often-cited argument against DB pensions is that individuals are retiring earlier and thus are spending a greater amount of time in retirement leisure collecting benefits—too much time from the point of view of those responsible for underfunded plans. While earlier retirement is one event potentially leading to more retirement leisure, what the argument ignores is that retirement leisure has two determinants that may be correlated: retirement age and death age. Ignoring one side of the retirement leisure determination may lead to misleading or even wholly inaccurate conclusions about who is consuming retirement leisure.

For example, research has shown that individuals who experience a major health shock retire earlier. If there is a positive correlation between retirement and death ages, and those individuals also die earlier, as could be expected, the earlier retirement ages may not translate into increased consumption of retirement leisure. The individual may even experience decreased retirement leisure if the mortality effect is strong enough. The opposite would occur if retirement and death are negatively correlated. If wealthier individuals retire earlier due to greater personal resources but also die later—both results that could be expected from existing research—an analysis based solely on retirement age would underestimate the true consumption of retirement leisure. The correlation between the two end points of retirement leisure leads to a flawed view of the retirement leisure distribution and necessitates a complete analysis in order to accurately guide retirement policy.

Results

We present the results from the ordered SUR model in Table 2. The coefficients and standard errors from the retirement age equation of the model are presented in the first two columns of Table 2, while the same information from the death age equation is presented in columns three and four. The final column presents the combined effect of the two determinants

on retirement leisure and is calculated by adding the negative of the retirement age coefficient (early retirement implies more retirement leisure) to the death age coefficient.

Although the primary contribution of the ordered SUR Tobit model is the joint retirement leisure estimation in column five, the individual equation results in columns one and three are comparable to existing research and can provide support for the validity of the model. Overall we find that our individual equation results match our expectations from prior research. Outside of demographic information, those in worse health retire significantly earlier as evidenced by a variety of different measures, as do individuals with greater personal wealth and who have lower DB pension accrual rates. There are not as many significant results for our mortality equation, but once again the coefficients match expectations, as women live over a year longer and those with various chronic health conditions such as cancer or diabetes die earlier. The individuals in our sample appear to behave similarly to those in other studies, and thus our analysis should be able to uncover common, potentially misleading retirement leisure predictions.

Examining retirement leisure jointly, we do find a number of factors that significantly affect the quantity of retirement leisure consumed. Being a woman leads to greater retirement leisure consumption both due to significantly earlier retirement and later death. While examining only retirement age does correctly identify that women consume significantly more retirement leisure, the magnitude is greatly underestimated by not adding the effect of later mortality. We find a number of other significant coefficients that also would have been correctly identified by examining only one side of the retirement leisure determination. These variables are significant in only one of the two equations, with the significant coefficient driving a significant joint effect as well. Individuals who are Hispanic, in good health, or have a DC pension consume over a year less retirement leisure due to later retirement ages with no offsetting effect from mortality. Particularly interesting is the result from the good health rating, as this variable induces later retirement without the expected boost in mortality. Arguments for raising retirement ages based on the fact that individuals are healthier at older ages and thus will live longer may not be valid and may result in decreasing the wellbeing of this group of retirees. Greater accrual rates from additional work and lesser DB value from prior jobs also significantly reduce retirement leisure by delaying retirement, but the effects are quite small. Having a stroke leads to over two and a half years more retirement leisure caused primarily by early retirement. While the early retirement behavior is expected given that less healthy people retire earlier, the fact that having a stroke does not affect mortality is somewhat unexpected.

One variable that deserves particular attention is the self-reported probability of living to age seventy-five relative to the life table probability con ditioned on age and sex. A value of one for the variable implies that the individual is as optimistic about living to age seventy-five as the life tables suggest he/she should be, while a value less/greater than one implies that the individual is less/more optimistic about living to age seventy-five than average. We include the life expectancy variable along with its square to test what Ghilarducci and Neuman (2004) term the "compensation hypothesis": that individuals may attempt to compensate for lower than average expected retirement leisure consumption due to low life expectancy by retiring earlier. The results do not support the compensation hypothesis because those individuals with average relative life expectancy consume over ten months less retirement leisure due primarily to later retirement. At low levels of relative life expectancy, individuals still consume over nine months less leisure due once again to primarily later retirement and not earlier death. If compensating for less expected retirement leisure is even a goal, individuals do not seem to be able to do so effectively.

The above results are interesting from the perspective of pension and quality of life research, but from an estimation strategy the results are somewhat unsurprising. Retiring significantly earlier and dying significantly later should lead to joint significance, while joint significance resulting from one significant individual coefficient essentially supports the argument for looking at only one determinant at a time. What is more interesting are those results that would be missed if only one side of the retirement leisure determination is examined. For example, individuals whose longest tenure position is in a low-skill, white collar occupation consume a year less retirement leisure than their counterparts in low-skill, blue collar occupations, even though neither of the individual equation coefficients is significant at the 5 percent level. Failing to take the correlation between the two events into consideration would completely miss the significant effect.

Some of the jointly insignificant coefficients deserve special attention as well. Individuals with arthritis both retire and die significantly earlier, with the joint effect for arthritic individuals being insignificant. Similarly, individuals with psychiatric problems and more financial wealth retire significantly earlier with no effect on mortality, but they do not consume significantly different amounts of retirement leisure. On the other hand, individuals with cancer, diabetes, or mobility problems die significantly earlier with no effect on retirement age, but they do not consume significantly less retirement leisure. Not only would the true net effect on leisure be missed in these situations, but the results from examining only one equation would lead to misleading results and badly designed public policy. Examining only the effect

on retirement age would suggest that those individuals with arthritis, psychiatric problems, or more financial wealth consumed more retirement leisure and thus would not be made worse off than the average individual by policies delaying their retirement. However, knowing that the net effect on retirement leisure is zero shows that delaying retirement for these groups would unambiguously make them worse off by reducing their retirement leisure consumption to below average levels.